7.4 Analytics Pane: Totals, References, Trends & Forecasts

7.4 Analytics Pane: Totals, References, Trends & Forecasts

The Analytics pane adds summaries and statistical models to a view. Every result depends on scope, grain, and retained marks. Treat these features as analytical evidence, not automatic proof of causation.

Totals and subtotals

Use Analysis > Totals or the Analytics pane to add row grand totals, column grand totals, and subtotals. Subtotals summarize intermediate levels when nested dimensions are present. Grand totals summarize the edge of the table.

The value shown at a total is not necessarily the arithmetic sum of visible cells. SUM is additive, but AVG, COUNTD, MEDIAN, and ratios may be recomputed at the total level. If category COUNTD values share customers, adding the displayed counts double-counts people. Choose the total aggregation behavior where Tableau offers it and verify a non-additive result independently.

Reference lines, bands, and distributions

A reference line marks a value such as an average, median, constant, parameter, or calculated measure. Its scope may be Table, Pane, or Cell. A table-scoped average uses all relevant marks in the table; a pane-scoped line is recomputed for each pane; a cell-scoped line applies to its cell.

A reference band shades the area between two values. A distribution band can show percentages, percentiles, quantiles, or standard-deviation regions depending on configuration. An average line is a reference-line shortcut for displaying the mean at the selected scope. These objects add context but do not filter records. Use a parameter-backed reference line when viewers should change a target interactively.

Box plots

A box plot summarizes a distribution with quartiles and a median. The box spans the first to third quartiles, and its interior line marks the median. Under the Data within 1.5 times the IQR option, the whiskers extend to the furthest observations within Q1 - 1.5 × IQR and Q3 + 1.5 × IQR. Observations beyond the whiskers appear as individual marks.

Those marks are outliers under this display rule; they are not automatically errors or statistically significant anomalies. Investigate source quality and business context before excluding them. Tableau can instead extend whiskers to the minimum and maximum when that option is selected.

Trend lines

Trend lines model relationships in views with supported continuous fields. Tableau offers model forms such as linear, logarithmic, exponential, power, and polynomial when the data satisfies their requirements. Show the model description to inspect coefficients, residual information, R-squared, and significance results.

R-squared describes how much observed variation the fitted model accounts for in the sample; it does not prove causation or guarantee useful predictions. A p-value is evaluated against a threshold chosen for the analysis. A conventional 0.05 threshold is common, but the threshold should be set before inspecting results and interpreted with sample size, assumptions, multiple testing, residuals, and business consequences. A p-value is not the probability that the relationship occurred by chance.

Trend models can be partitioned by color or other dimensions, producing separate models. Confirm whether the question requires one model across the table or models per category. Excluding marks changes the fitted population.

Forecasting

Tableau's built-in forecast uses exponential-smoothing models. A view needs a supported date field or an ordered dimension and at least one measure. A model needs sufficient observations: at least five data points for estimating a trend, and seasonality generally needs at least two seasons or one season plus five observations. More data may be needed for a stable business forecast.

Forecasting can be unavailable when the view contains unsupported constructs such as table calculations, percent-of-total measures, disaggregated measures, or certain date configurations. Diagnose the shelves before blaming live versus extract mode.

Automatic model selection compares supported trend and season combinations. Additive seasonality models a seasonal effect with roughly constant absolute magnitude; multiplicative seasonality models an effect that scales with the series level. Customize forecast length, prediction interval, seasonality, trend, and ignored recent periods when the data and use case justify it. By default Tableau can ignore the latest partial period; verify whether that period is actually incomplete.

A prediction interval expresses model uncertainty under its assumptions. It is not a guarantee that a future value will fall inside the band. Review residuals, structural breaks, missing periods, and whether the forecast horizon exceeds what the history can support.

Explain Data

Explain Data can examine a selected mark and surface supported statistical explanations, including candidate fields not currently displayed. Tableau excludes unsupported or unsuitable fields and authors can control which fields are eligible. The results identify associations or unusual contributing records for investigation; they do not establish a root cause.

Use Explain Data as a starting point. Check data quality, grain, filters, and alternative hypotheses. A candidate explanation can be affected by correlated fields, small groups, or selection choices.

Choosing the right object

Verification checklist

1. Confirm the grain and retained population.
2. Set Table, Pane, or Cell scope deliberately.
3. Reconcile totals, especially for non-additive measures.
4. Inspect model assumptions and residuals rather than relying on one statistic.
5. Verify enough ordered history exists for the forecast components.
6. Describe associations and uncertainty without claiming causation.

An analytical overlay is only as sound as the population, scope, and assumptions supplied to it.